0
$\begingroup$

I am new to ML and was exploring a time-series dataset for the very first time. The aim was to predict the volume of vehicles passing one of the 4 junctions given some historical data. After dividing the DateTime column into separate columns like (Year,Month,Day etc), I started performing visualizations to see the trend of target variable(volume of vehicles in a junction) in this case and found out that the mean volume of vehicles more or less remain the same over some of the features. Please refer image below(Mean of vehicles passing 4 junctions/day): enter image description here My question is that if we observe that a particular feature is not effecting the target variable much, can we directly assume that this feature will not be helpful to a ML model(as my intuition suggests) or am I completely wrong with my thinking?

$\endgroup$
1
$\begingroup$

There are cases each feature doesn't correlate with the target variable but the combination of features is strongly correlated with the target variable.

In the extreme, there is the famous XOR problem. Let's say there are two features and one target variable. Two features and the target variable are binary taking True or False. And the target variable is decided by the XOR relationship of two features.

In this setting, when feature 1 takes True(or False), there is a 50% chance for the target variable is True or False. (same for the feature 2). That is to say, correlation with one feature and target is zero. But when we consider the correlation of both features 1 and 2 with the target variable, the correlation is 1.

You better not ignore a feature from the independence to the target variable because a combined feature may be correlated with the target.

We can do the subset selection instead of considering one feature at a time. We consider all the subset of features. And do the experiment on each of the subsets and choose the best subset of features. But this takes a long time as there are $2^p$ possibilities (p: number of features). So they start to develop more efficient algorithms to find an approximately best subset usually using a greedy algorithm.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.